Ten papers covering retrieval, RAG eval, and factuality scoring. BEIR is the retrieval baseline that survives. FActScore is the atomic-fact pattern that everything else borrows from. RAGAS, Self-RAG, and ARES wire those metrics into pipelines; the RAGAS library docs for v0.4 list evaluate() as deprecated in favor of the experiment decorator (checked as of October 2026). SAFE is the search-augmented evaluator result that makes large-scale factuality scoring cheap. SimpleQA, its cleaned 1,000-prompt successor SimpleQA Verified, and HalluLens are the work to read before claiming a hallucination rate. HalluHard (2026) is the one to read before you trust a single-turn number: it scores multi-turn answers that carry inline citations, checks each citation against fetched sources, and found about 30% hallucination even for the strongest setup it tested, Opus 4.5 with web search.